NFS Problem

Jakob Oestergaard jakob at unthought.net
Fri Oct 3 01:13:37 EDT 2003


On Thu, Oct 02, 2003 at 08:59:55PM -0400, Hao He wrote:
> Hi, there.
> 
> I am building  a cluster with 16 or 32 nodes, based on Pentium 4, Intel 875P
> chipsets and Intel CSA Gigabit NIC.
> The distribution is RedHat 9.
> I have some experience before but I still got some problem in NFS.
> 
> Problem 1: When I just use 'rw' and 'intr' as the parameters used in
> /etc/fstab, I got following problem when startup clients (while the server
> with NFS daemon is running):
> Mount: RPC: Remote system error -- no route to host

That's a network problem or a network configuration problem.

Usually this would be a name resolution problem. Check that the hostname
in your fstab can be resolved in early boot (add it to your hosts file
if necessary), or use the IP address of the server instead.

But the error message seems to indicate that it's not resolution but
routing - very odd...  Is the network up?   Do you have any special
networking setup?

Try checking your init-scripts to see that the network is really started
before the NFS filesystems are mounted.

> Then I added 'bg' to /etc/fstab, this time the result is better. Several
> minutes after the client booted up, the remote directory mounted.

So you NFS mount depends on something (network related) that isn't up at
the time when the system tries to mount your NFS filesystems.

Either you have a special (and wrong) setup, or RedHat messed up good :)

Check the order in which things are started in your /etc/rc3.d/
directory.  Network should go before NFS.

> However, in many cases following meassage was prompted:
> nfs warning: mount version older than kernel

Most likely this is not really a problem - I've had systems with that
message work just fine.

You could check to see if RedHat has updates to mount.

> 
> Problem 2: I am mounting two remote directories from the server, however, at
> some nodes, only one directory even no directory got mounted.
> If only one directory mounted successfully, it differs from one client to
> another, and to the same node, it changes from time to time at system
> booting up, like dicing.
> This really confused me.

Isn't this problem 1 over again?

> 
> Problem 3: Sometimes I got the message at the server node like this:
> (scsi 0:A:0:0): Locking max tag count at 33.

That's a SCSI diagnostic. You can ignore it.

> However, seems it does not make trouble to mounted directories.
> I think it must be related with NFS.

It's not related to NFS.

> 
> I have a further question: Since there may be 16 or 32 or even more clients
> try to mount the remote directory at the same time,
> can the NFS server really handle so much requests simultaneously? Is there
> any effective alternate method to share data, besides NFS?

That should be no problem at all.

NFS should be up to the task with no special tuning at all.

Once you have all your nodes mounting NFS properly, you can start
looking into tuning for performance - but it really should work 'out of
the box' with no special tweaking.

> 
> How to solve these problems? Any suggestion?
> Thank you very much. I will appreciate your response.

Use the following options to the NFS mounts in your fstab:
  hard,intr

You can add
  rsize=8192,wsize=8192
for tuning.

You should not need 'bg' - although it may be convenient if you need to
be able to boot your nodes when the NFS server is down.

One thing you should make sure: never use host-names or netgroups in
your exports file on the server (!)   *Only* use IP addresses or
wildcards - *Never* use names.  Using names in your 'exports' file on
the server can cause *all* kinds of weird sporadic irreproducible
problems - it's a long-standing and extremely annoying problem, but
fortunately one that has an easy workaround.

Check:
*) Server: Your exports file (only IP or wildcard exports)
*) Clients: Your fstab (use server IP or name in hosts file)
*) Clients: Is network started before NFS mount?

Please write to the list about your progress  :)

-- 
................................................................
:   jakob at unthought.net   : And I see the elder races,         :
:.........................: putrid forms of man                :
:   Jakob Østergaard      : See him rise and claim the earth,  :
:        OZ9ABN           : his downfall is at hand.           :
:.........................:............{Konkhra}...............:
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list